igergov.com

Publications

Journal and conference papers on IoT, artificial intelligence, critical infrastructure and security.

5journal articles
3conference papers
8 of 8

2026

  1. published journal Traffic Optimization with Smart Traffic Lights: An Overview of Algorithms and Implementations Ivo Gergov, Rumen Doynov Proceedings of the Technical University of Sofia · Technical University of Sofia

    A review of the algorithms and real-world deployments of smart traffic lights between 2015 and 2025: from centralised optimisation and max-pressure to single- and multi-agent reinforcement learning. It looks at edge computing, V2X communication and the move from simulation to a real city, and orders the approaches by maturity. It is useful for cities looking for ways to cut congestion and emissions with the data and infrastructure they already have.

    doi:10.47978/tus.2026.76.03.012 Article →
  2. published journal Large Language Models for Recovery Plan Generation in Internet-Connected Critical Infrastructures: Architectures, Applications, Limitations, and Research Directions Georgi Tsochev, Ivo Gergov Future Internet · 18, 295 · MDPI

    How large language models can help restore critical infrastructure after a cyberattack: energy, water, transport, hospitals. The review orders the approaches from a simple assistant to multi-agent systems with verification and proposes a reference architecture. The conclusion is that assistants which rely on real data and leave the decision to a person make sense today, while autonomous recovery is still premature. It matters at a time when language models are entering operational work faster than ways to trust them appear.

    doi:10.3390/fi18060295 Article →
  3. published journal Multi-Agent Reinforcement Learning for Decentralized Computational Resource Negotiation in a Tokenized Double-Auction Market: A Reproducibility-First Benchmark Ivo Gergov, Georgi Tsochev, Adelina Aleksieva-Petrova Mathematics · 14, 1828 · MDPI

    A market where machines sell their spare computing power and buy what they need through double-auction trading with tokens. Four agent-training algorithms (PPO, MAPPO, MADDPG, IQL) are compared in one and the same environment, with 12 runs and confidence intervals, so that every result can be reproduced. The work comes at a time when computing power is expensive and distributed, and comparisons between such algorithms can rarely be repeated.

    doi:10.3390/math14111828 Article →
  4. published journal Critical Infrastructure Restoration and Artificial Intelligence Systems: Applications and Practical Limitations Ivo Gergov, Maksim Sharabov, Alexander Rusev, Georgi Tsochev Sustainability · 18, 5297 · MDPI

    A systematic review following PRISMA 2020 of how artificial intelligence helps restore critical infrastructure after a disaster or an outage. It covers 53 sources, scientific and official: the NIS2 and CER directives, the AI Act, continuity practice, SCADA, digital twins and model families. It shows that AI brings real benefit when it is built into interoperable systems under human oversight and within regulation, while many of the applications have so far been tested only in simulation. It arrives in the year the new European rules on the resilience of critical entities start to apply.

    doi:10.3390/su18115297 Article →

2025

  1. published conference Security Vulnerabilities of Smart Home: A Case Study of Smart Camera Ivo Gergov, Georgi Tsochev, Maria Nenova TELECOM 2025 · IEEE

    An analysis of the security problems of smart home cameras and of how well they are protected. The camera is one of the devices that sees the most private space, and it is often connected to the cloud and to the other devices in the home. This work is part of the reason I keep as much as possible local in my own home.

    doi:10.1109/TELECOM66943.2025.11304063 Article →
  2. published journal Watermarking Fine-Tuning Datasets for Robust Provenance Ivo Gergov, Georgi Tsochev Applied Sciences · 15, 10457 · MDPI

    A method for hiding a watermark in a fine-tuning dataset for a language model, so that the owner can later check whether a model was trained on their data. The mark is embedded by choosing synonyms with a secret key and detected with a statistical test. The paper also reports the negative results honestly: on short texts and with models such as Mistral 7B detection is weak, and the limitations are described in detail. It matters at a time when companies fine-tune models on their own corpora and the question of where the data came from becomes a legal one.

    doi:10.3390/app151910457 Article →

2023

  1. published conference Exploring the Role of Chat Bots in the Modern World: Architecture, Types, and Areas of Application Ivo Gergov, Ognyan Nakov, Nadezhda Angelova, Irina Ivanova COMSCI 2023 · IEEE

    A review of chatbots: their architecture, the kinds that exist and where they are used, including in the smart home. It was published in 2023, the year chatbots went mainstream.

    doi:10.1109/COMSCI59259.2023.10315893
  2. published conference Zero-Trust in Instant Messaging Ivo Gergov, Nikola Obreshkov COMSCI 2023 · IEEE

    Applying the Zero Trust model to instant messaging apps: no device and no connection is trusted by default. The paper shows that this significantly reduces the risk of data leaks and unauthorised access. It was written at a time when leaking and selling personal data had become a problem all over the world.

    doi:10.1109/COMSCI59259.2023.10315901